Funded by the European Union. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Education and Culture Executive Agency (EACEA). Neither the European Union nor EACEA can be held responsible for them. Project number: 2023-1-NL01-KA220-HED-000155675. Casus 31: Ondersteuning van collaboratief online wetenschapsonderwijs met een overdraagbare en configureerbare conversatie-agent General information Reference/Source: Araujo, A. De, Papadopoulos, P. M., McKenney, S., & Jong, T. De. (2023). Supporting Collaborative Online Science Education with a Transferable and Configurable Conversational Agent. Computer-Supported Collaborative Learning Conference, CSCL, 2023-June, 416–419. https://doi.org/10.22318/cscl2023.469853 Institution: University of Twente (Netherlands) Course/Subject: Science, photosynthesis Aim: To develop and pilot a transferable and configurable conversational agent (Clair) designed to facilitate productive talk in collaborative online learning environments. Target group: Students in pairs within collaborative online learning settings (AI developed to support students learning process) Description of case Overview: Researchers designed a conversational agent named Clair to foster productive talk in collaborative online learning environments. Clair is intended to be transferable to different topics and languages and allow for a degree of teacher configuration. Intervention: The pilot study used a within-subjects experiment. Students worked in pairs on a Go-Lab activity about photosynthesis. After an initial phase without Clair, dyads were assigned to 'control' or 'treatment' groups, with the treatment group receiving Clair's interventions. Clair's design: Clair used talk moves (e.g., Add-on, Rephrasing, Expand Reasoning) to stimulate discussion based on a combination of dialogue variables (focus, intent, topic similarity, etc.) and fuzzy logic rules. Lessons learned Limited impact: While Clair showed some potential in increasing explicit reasoning and decreasing participation imbalance, the overall effect wasn't statistically significant. Design issues: Clair's interventions were perceived as repetitive and robotic. The triggering mechanisms and rules could be improved. Unrealistic expectations: Students expected Clair to provide more direct content support, which isn't its intended function. Key takeaways: Designing an effective conversational agent for collaborative learning is complex. Future iterations should focus on more nuanced interventions, better rule design, and managing student expectations about the CA's capabilities. Implications for practice N/A
RkJQdWJsaXNoZXIy NzYwNDE=